On Mon, 2023-03-20 at 08:17 +0000, David Woodhouse wrote:
> 
> In _cpu_up() the call to idle_thread_get() only happens if the CPU in
> question is starting from CPUHP_OFFLINE but I think that's sufficient?
> 
> diff --git a/kernel/cpu.c b/kernel/cpu.c
> index 6c0a92ca6bb5..43e0a77f21e8 100644
> --- a/kernel/cpu.c
> +++ b/kernel/cpu.c
> @@ -591,12 +591,6 @@ static int bringup_cpu(unsigned int cpu)
>  	struct task_struct *idle = idle_thread_get(cpu);
>  	int ret;
>  
> -	/*
> -	 * Reset stale stack state from the last time this CPU was online.
> -	 */
> -	scs_task_reset(idle);
> -	kasan_unpoison_task_stack(idle);
> -
>  	/*
>  	 * Some architectures have to walk the irq descriptors to
>  	 * setup the vector space for the cpu which comes online.
> @@ -1383,6 +1377,12 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
>  			ret = PTR_ERR(idle);
>  			goto out;
>  		}
> +
> +		/*
> +		 * Reset stale stack state from the last time this CPU was online.
> +		 */
> +		scs_task_reset(idle);
> +		kasan_unpoison_task_stack(idle);
>  	}
>  
>  	cpuhp_tasks_frozen = tasks_frozen;

I put that into a separate commit, rebased on tip/x86/apic and pushed
it out with the rest of the changes to
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/parallel-6.2-rc3-v16

I'll let Usama do another round of testing and repost it (please).

Thanks!